This LoRA was trained with CivitAI's trainer. The captioning is 100% natural language, captioned by the default CivitAI captioner (I am not sure which one it is).

The model, at strength 1.0, often forces a "from above", high angle perspective due to the bias of the training data. To prevent it from doing this, if you don't like or want it, lowering the strength to as far as 0.8 should help.